Physics-Constrained Joint Inversion for Reservoir Fluid Mapping
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing reservoir monitoring methods are insufficient for comprehensive mapping of fluid distribution in the interwell space, relying on insufficient well patterns and remote sensing techniques that fail to accurately predict fluid movements due to heterogeneous rock formations.
Innovation Solution
A hybrid scheme of physics-driven inversion and statistical deep learning inversion is implemented, combining physics-based and data-driven approaches through a reciprocal feedback loop to optimize the estimation of multiple model parameters, using a deep learning neural network trained with examples to predict parameter distributions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If traditional remote sensing techniques are used for reservoir monitoring, then the measurement coverage is limited, but the measurement precision of fluid distribution is insufficient
Solution Approach 1:
The patent combines physics-based inversion methods with deep learning-based inversion methods into a unified hybrid framework. The physics-based component ensures measurement coverage and physical consistency, while the deep learning component enhances prediction accuracy for fluid distribution in heterogeneous rock formations, resolving the contradiction between coverage and precision.
Solution Approach 2:
The invention creates a composite inversion approach that integrates two distinct methodologies (physics-based and data-driven deep learning) into a single hybrid system. This composite approach leverages the strengths of both methods to achieve both broad measurement coverage and high prediction accuracy simultaneously.
2Area of stationary object
If well patterns are increased to improve fluid distribution mapping, then the measurement coverage is improved, but the device complexity and cost increase
Solution Approach 1:
The patent replaces the mechanical approach of increasing well infrastructure with an intelligent software-based hybrid inversion system. Instead of physically adding more wells to cover interwell spaces, the system uses combined physics-based and deep learning-based inversion algorithms to accurately map fluid distributions using existing well data, reducing device complexity while improving coverage.
3Stability of the object's composition
If physics-based inversion is used alone, then the physical consistency is maintained, but the prediction accuracy in heterogeneous formations is insufficient
Solution Approach 1:
The hybrid inversion framework merges physics-based inversion (which maintains physical consistency through governing equations) with deep learning-based inversion (which captures complex patterns in heterogeneous formations from training data). The coupling operator integrates both approaches, allowing the system to maintain physical consistency while achieving high prediction accuracy in heterogeneous rock formations.
4Measurement precision
If deep learning inversion is used alone, then the prediction accuracy is improved, but the physical consistency may be compromised
Solution Approach 1:
The coupling operator in the hybrid framework combines deep learning predictions with physics-based constraints. The deep learning component provides high prediction accuracy for parameter distributions, while the physics-based component ensures the results satisfy physical laws and conservation principles, maintaining physical consistency even in complex heterogeneous formations.
Data Source
AI summary
A deep learning framework includes a first model for predicting one or more attributes of a system; a second model for predicting one or more attributes of the system; at least one coupling operator combining the first and second models; and at least one inversion module for receiving the combined first and second models from the coupling operator. The inversion module simultaneously optimizes the first model and the second model, thereby resulting in a composite objective function representative of a prediction that is outputted to at least one user.